aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorSiri Hansen <[email protected]>2013-03-25 14:46:59 +0100
committerSiri Hansen <[email protected]>2013-03-25 14:46:59 +0100
commitcea6232b9f5c44f76ff1023aa78ea603cb4ed0d9 (patch)
tree8c78b85dfbd9ef2c17693823155efb207deade63
parent19ee613287ae423a09daf448e3b9bde96e1c710b (diff)
downloadotp-cea6232b9f5c44f76ff1023aa78ea603cb4ed0d9.tar.gz
otp-cea6232b9f5c44f76ff1023aa78ea603cb4ed0d9.tar.bz2
otp-cea6232b9f5c44f76ff1023aa78ea603cb4ed0d9.zip
[common_test] Add support for testing repeated testruns
ct_test_support which is used in common_test's own tests did not support testruns with -repeat, -duration or -until options. This is now added.
-rw-r--r--lib/common_test/test/ct_test_support.erl70
1 files changed, 46 insertions, 24 deletions
diff --git a/lib/common_test/test/ct_test_support.erl b/lib/common_test/test/ct_test_support.erl
index 5e109e98e9..70dd087358 100644
--- a/lib/common_test/test/ct_test_support.erl
+++ b/lib/common_test/test/ct_test_support.erl
@@ -29,7 +29,8 @@
-export([init_per_suite/1, init_per_suite/2, end_per_suite/1,
init_per_testcase/2, end_per_testcase/2,
write_testspec/2, write_testspec/3,
- run/2, run/3, run/4, get_opts/1, wait_for_ct_stop/1]).
+ run/2, run/3, run/4, run_ct_run_test/2, run_ct_script_start/2,
+ get_opts/1, wait_for_ct_stop/1]).
-export([handle_event/2, start_event_receiver/1, get_events/2,
verify_events/3, verify_events/4, reformat/2, log_events/4,
@@ -224,9 +225,15 @@ get_opts(Config) ->
%%%-----------------------------------------------------------------
%%%
run(Opts, Config) when is_list(Opts) ->
+ %% use ct interface
+ CtRunTestResult=run_ct_run_test(Opts,Config),
+ %% use run_test interface (simulated)
+ ExitStatus=run_ct_script_start(Opts,Config),
+ check_result(CtRunTestResult,ExitStatus,Opts).
+
+run_ct_run_test(Opts,Config) ->
CTNode = proplists:get_value(ct_node, Config),
Level = proplists:get_value(trace_level, Config),
- %% use ct interface
test_server:format(Level, "~n[RUN #1] Calling ct:run_test(~p) on ~p~n",
[Opts, CTNode]),
CtRunTestResult = rpc:call(CTNode, ct, run_test, [Opts]),
@@ -242,7 +249,11 @@ run(Opts, Config) when is_list(Opts) ->
timer:sleep(5000),
undefined = rpc:call(CTNode, erlang, whereis, [ct_util_server])
end,
- %% use run_test interface (simulated)
+ CtRunTestResult.
+
+run_ct_script_start(Opts, Config) ->
+ CTNode = proplists:get_value(ct_node, Config),
+ Level = proplists:get_value(trace_level, Config),
Opts1 = [{halt_with,{?MODULE,ct_test_halt}} | Opts],
test_server:format(Level, "Saving start opts on ~p: ~p~n",
[CTNode, Opts1]),
@@ -253,27 +264,38 @@ run(Opts, Config) when is_list(Opts) ->
ExitStatus = rpc:call(CTNode, ct_run, script_start, []),
test_server:format(Level, "[RUN #2] Got exit status value ~p~n",
[ExitStatus]),
- case {CtRunTestResult,ExitStatus} of
- {{_Ok,Failed,{_UserSkipped,_AutoSkipped}},1} when Failed > 0 ->
- ok;
- {{_Ok,0,{_UserSkipped,AutoSkipped}},ExitStatus} when AutoSkipped > 0 ->
- case proplists:get_value(exit_status, Opts1) of
- ignore_config when ExitStatus == 1 ->
- {error,{wrong_exit_status,ExitStatus}};
- _ ->
- ok
- end;
- {{error,_}=Error,ExitStatus} ->
- if ExitStatus /= 2 ->
- {error,{wrong_exit_status,ExitStatus}};
- ExitStatus == 2 ->
- Error
- end;
- {{_Ok,0,{_UserSkipped,_AutoSkipped}},0} ->
- ok;
- Unexpected ->
- {error,{unexpected_return_value,Unexpected}}
- end.
+ ExitStatus.
+
+check_result({_Ok,Failed,{_UserSkipped,_AutoSkipped}},1,_Opts)
+ when Failed > 0 ->
+ ok;
+check_result({_Ok,0,{_UserSkipped,AutoSkipped}},ExitStatus,Opts)
+ when AutoSkipped > 0 ->
+ case proplists:get_value(exit_status, Opts) of
+ ignore_config when ExitStatus == 1 ->
+ {error,{wrong_exit_status,ExitStatus}};
+ _ ->
+ ok
+ end;
+check_result({error,_}=Error,2,_Opts) ->
+ Error;
+check_result({error,_},ExitStatus,_Opts) ->
+ {error,{wrong_exit_status,ExitStatus}};
+check_result({_Ok,0,{_UserSkipped,_AutoSkipped}},0,_Opts) ->
+ ok;
+check_result(CtRunTestResult,ExitStatus,Opts)
+ when is_list(CtRunTestResult) -> % repeated testruns
+ try check_result(sum_testruns(CtRunTestResult,0,0,0,0),ExitStatus,Opts)
+ catch _:_ ->
+ {error,{unexpected_return_value,{CtRunTestResult,ExitStatus}}}
+ end;
+check_result(CtRunTestResult,ExitStatus,_Opts) ->
+ {error,{unexpected_return_value,{CtRunTestResult,ExitStatus}}}.
+
+sum_testruns([{O,F,{US,AS}}|T],Ok,Failed,UserSkipped,AutoSkipped) ->
+ sum_testruns(T,Ok+O,Failed+F,UserSkipped+US,AutoSkipped+AS);
+sum_testruns([],Ok,Failed,UserSkipped,AutoSkipped) ->
+ {Ok,Failed,{UserSkipped,AutoSkipped}}.
run(M, F, A, Config) ->
run({M,F,A}, [], Config).